Fig. 5. Prediction of enantioselectivity by SEMG-MIGNN model (steric- and electronics-embedded molecular graph with molecular interaction graph neural network).

a Overview of the dataset of the chiral phosphoric acid-catalyzed thiol addition to N-acylimines. These data are originally published by Denmark and co-workers²⁶. b Regression performances of baseline MG-GCN (baseline molecular graph and graph convolutional network), SEMG-GCN (steric- and electronics-embedded molecular graph with graph convolutional network), baseline MG-MIGNN (baseline molecular graph with molecular interaction graph neural network) and SEMG-MIGNN models (steric- and electronics-embedded molecular graph with molecular interaction graph neural network). The dataset was randomly split into 600 (training) and 475 (test) reactions. SEMG-MIGNN model (steric- and electronics-embedded molecular graph with molecular interaction graph neural network) outperforms the other tested combinations with R² and RMSE (root mean square error) of 0.915 and 0.197 kcal mol⁻¹ respectively. Source data are provided as data2.csv and Data_for_Fig_5.csv.